Synchronous baseband digital optical communication systems require low-pass electronic filters within transmitters for spectral shaping, and within receivers for limitation of noise bandwidth. Low pass filters having Gaussian characteristics would be desirable for these applications because they provide amplitude and phase response characteristics which minimize the distortion imparted to a pulse code modulated signal. Additionally, high bit rate SONET systems employing duobinary coding of optical phase, for example, systems employing the OC-192 protocol, may require a band-limiting filter within the duobinary encoder. Such a filter should exhibit Gaussian characteristics to minimize pulse distortion, low loss to enhance efficiency, and low reflection since the signal bandwidth entering the filter is typically twice the cut-off frequency.
Conventional low pass Gaussian filters are unable to meet the specifications dictated by the above applications because they suffer from a high voltage standing wave ratio (VSWR), i.e., high reflection. When transmission bit rates extend into microwave frequencies, reflections and the resulting standing waves generate distortion which can severely impair the communication system.
Conventional low pass, low reflection (absorptive) microwave filters are also unable to meet the specifications dictated by the above applications. Conventional low pass, low reflection microwave filters are constructed of singly terminated lossless low pass and high pass networks connected to form a diplexer. FIG. 1 illustrates the generic form of such a filter. As shown in FIG. 1, the load is connected to port 2-2' and the source is connected to port 1-1', each of which has a 50.OMEGA. real internal impedance. The high pass network absorbs power in the stop band of the low pass network to minimize reflection at the driven port 1-1'. The frequency response of the filter is determined by the frequency responses of the low pass and high pass networks. For maximally flat attenuation and Tchebyscheff filter classes, the low pass and high pass networks can be implemented using standard synthesis techniques.
It is not possible to implement Gaussian filters having maximally flat time-delay using diplexer structures because phase ripple at the crossover frequency between the low pass and high pass networks generates signal distortion. Diplexer configurations also have the disadvantage of being nonreciprocal. That is, exchanging the source and load ports does not yield the same transfer characteristic from source to load. In addition, the reflection coefficient at port 2-2' is significantly greater than that at port 1-1' since port 2-2' is not diplexed. This yields a transfer characteristic that is highly sensitive to load matching at port 2-2'. Diplexing port 2-2' would alleviate this problem, but at the expense of severely impacting the transfer characteristic.
Accordingly, conventional filter technology does not enable the implementation of a low pass filter exhibiting low loss, low reflection, and Gaussian frequency response characteristics at microwave and radio frequencies.